WESWind Energy ScienceWESWind Energ. Sci.2366-7451Copernicus PublicationsGöttingen, Germany10.5194/wes-2-229-2017Field test of wake steering at an offshore wind farmFlemingPaulpaul.fleming@nrel.govhttps://orcid.org/0000-0001-8249-2544AnnoniJenniferShahJigar J.WangLinpengAnanthanShreyasZhangZhijunHutchingsKyleWangPengChenWeiguoChenLinNational Wind Technology Center, National Renewable Energy Laboratory, Golden, CO 80401, USAResearch & Development, Envision Energy USA Ltd, Houston, TX 77002, USAResearch & Development, Envision Energy Ltd, Shanghai, 200051, ChinaPaul Fleming (paul.fleming@nrel.gov)8May20172122923916January20176February20174April20177April2017This work is licensed under a Creative Commons Attribution 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/This article is available from https://wes.copernicus.org/articles/2/229/2017/wes-2-229-2017.htmlThe full text article is available as a PDF file from https://wes.copernicus.org/articles/2/229/2017/wes-2-229-2017.pdf
In this paper, a field test of wake-steering control is
presented. The field test is the result of a collaboration between the
National Renewable Energy Laboratory (NREL) and Envision Energy, a smart
energy management company and turbine manufacturer. In the campaign, an array
of turbines within an operating commercial offshore wind farm in China have
the normal yaw controller modified to implement wake steering according to a
yaw control strategy. The strategy was designed using NREL wind farm models,
including a computational fluid dynamics model, Simulator fOr Wind Farm Applications (SOWFA), for understanding wake
dynamics and an engineering model, FLOw Redirection and Induction in Steady State (FLORIS), for yaw control optimization.
Results indicate that, within the certainty afforded by the data, the
wake-steering controller was successful in increasing power capture, by
amounts similar to those predicted from the models.
Introduction
Wind farm control is an active field of research in which the controls of
individual turbines co-located within a wind farm are coordinated to improve
the overall performance of the farm. One objective of wind farm control is
improving the power production of wind farms by accounting for the wake
interactions between nearby turbines.
In one wind farm control concept, turbines are yawed to introduce a
deflection of the wake away from downstream turbines. This method has been
referred to as “controlling the wind” and
“yaw-based wake steering” . High-fidelity
simulations of wake steering have shown the potential of this technique.
used computational fluid dynamics (CFD)
simulations to demonstrate the wake deflection capability of wind turbines
and provided a model of this deflection. In ,
they used National Renewable Energy Laboratory (NREL)'s CFD-based Simulator fOr Wind Farm Applications (SOWFA) to
investigate the capabilities of wind turbines to redirect wakes. In
, the behavior of wake steering in different
atmospheric conditions was investigated, also using CFD. Finally, in
, simulations of two-turbine wind farms, again
using SOWFA, were used to show that through wake steering, the net power of
the two turbines is increased when the upstream turbine applies an
intentional yaw misalignment.
Based on high-fidelity simulations, there appears to be good opportunities
for improved power performance of wind farms with significant wake losses.
Recent efforts have focused on the design of lower-fidelity,
controller-oriented models, and controllers based on these models that use
wake steering to actively improve power. In , the FLOw
Redirection and Induction in Steady State (FLORIS) model is described and
used to determine optimal yaw settings for a model six-turbine wind farm. Set
points for a particular wind speed and direction are determined by optimizing
the yaw angles of the turbines using FLORIS, and these set points are used in
SOWFA simulations. The results from SOWFA agree with the predictions from
FLORIS, and total power capture is increased by 13 %. This work is carried
further, and FLORIS is used to assess the overall improvement from control,
first for one speed and over a wind rose of directions in
and then to determine the overall annual energy
production in . These studies indicate a good
potential for improved overall annual power production for wind farms
experiencing significant wake losses. It should be noted that even greater
benefits can be yielded if future wind farms are designed for active control
of wakes, rather than using large inter-turbine spacings to avoid wake
losses. This combined optimization of wind farm control and wind farm system
engineering is a subject of active research. These simulation studies have
demonstrated a theoretical potential of wind farm control. However, it is
often noted that issues arising in implementation in real conditions might
undermine the positive results. Inaccuracies in the control-oriented models
or high-fidelity simulations have been cited as a potential issue. Additional
modeling of constantly changing wind direction could improve the comparison
between simulation and field testing.
Some field testing of wake steering has been performed to date to understand
the potential of this wind farm control strategy. In
, wake steering is implemented at a scaled
wind farm; however, the results are inconclusive. Wind tunnel testing of wake
steering is performed in and
, and the results are encouraging because in each
case that wake steering is observed, overall power capture is improved for
two-turbine cases. The results are in alignment with earlier simulation
studies conducted by , ,
and in the amount of improvement and the
asymmetric relationship of yaw misalignment and power improvement. On the
full scale, a nacelle-mounted lidar is used to observe wake deflection on a
utility-scale turbine in . Finally, in
, two ongoing studies of wake steering are
presented, which are both part of a multiyear US Department of Energy
Atmosphere to Electrons project. In one of the field studies at the Scaled
Wind Farm Test Facility (SWiFT), two 27 m diameter (D) turbines
(intentionally aligned in the dominant wind direction at a spacing of 5 D) were used to perform a comprehensive test of wake steering. A
second ongoing experiment was done, in which a rear-mounted lidar was used to
monitor a wake of a utility-scale turbine, which is set to hold a specific
yaw misalignment for a prolonged period of time. Initial results from both
campaigns are in accord with the model predictions described earlier.
In the present study, a wind farm controller that performs wake steering is
designed and implemented for an operating commercial offshore wind farm in
China with Envision turbines. The control strategy is simple and based on the
approach used to control the simulated wind farm in .
The test was run for several months and data were collected and compared from
time periods when the controller was operating and not operating. The results
and analysis indicate a successful improvement in power production.
Additionally, the data provide important validation of the models,
specifically SOWFA and FLORIS, used in the design of the controller, and can
be generalized to similar CFD and control-oriented models. There are some
qualifications to these results that will be fully considered in the text.
The contributions of this paper include the results and analysis from a wind
farm test of wake-steering control. The project is a collaborative project
between the NREL and Envision Energy,
a smart energy management company and turbine manufacturer. The positive
results motivate further efforts into the design and development of such
control. Additionally, this paper provides evaluation on the performance of
wind farm control modeling tools in their ability to predict the effect of
wind farm control strategies.
Project overview
The goal of the project was to implement a test of yaw-based wake steering at
an operating wind farm. The project was broken down into a workflow
illustrated in Fig. , which shows the stages of work and
the structure by which this paper is organized.
This figure demonstrates the workflow of this project. In
particular, this project started by running CFD models of the turbines in
this field study to understand the wake characteristics. The FLORIS model was
tuned to those CFD simulations. A control look-up table was generated that
contained the optimal yaw settings for the control turbine. A field test was
conducted with the controller on and off. The data were analyzed and the
results are presented in this paper. LUT – look-up table; SCADA –
supervisory control and data acquisition.
(a) Rudong wind farm used in the field study. (b)
Turbine locations. The wake-steering control strategy is implemented to
mitigate the wake interactions between C1, on the one hand, and D1, D2, and D3, on the other. As indicated
by the diagram, C1 is the control turbine, R1 is the reference turbine, and
D1, D2, and D3 are downstream turbines waked by C1.
Turbines O1 and O2 are other turbines not directly used in the study but
whose wakes are noticed in certain directions.
The first stage of work is the selection of a wind farm for use in the
experiment. The Longyuan Rudong Chaojiandai offshore wind farm in Jiangsu,
China, consists of turbines from multiple manufacturers undergoing various
phases of construction. A selected portion of the site was studied for this
effort, consisting of 25 Envision EN136/4 MW turbines incorporating a
high-speed three-stage gearbox and induction generator (see
Fig. a).
From the wind farm, a subset of turbines was selected for implementing the
experiment. These turbines form the front two rows of turbines for winds
coming from the northeast. The arrangement of these turbines and their names
to be used throughout this paper are shown in Fig. .
For the campaign, a single turbine was selected to be the controlled turbine.
This is turbine C1, indicated in Fig. . The turbine is shown to
wake three particular turbines: D1, from a wind direction of 340∘
at a distance of 7 D; D2, from a wind direction of 51∘ at a distance
of 8.6 D; and D3, from a wind direction of 81∘ at a distance of
14.3 D. A control strategy was designed to improve the summed power of turbine
C1 and a downstream turbine (i.e., C1 and D1, C1 and D2, and C1 and D3) through yaw misalignment of turbine C1. This approach provides three tests of
wake steering at different inter-turbine distances. Of these, the pairing with
turbine D1, at 7 D, is most promising. This is the distance used in
, which showed a potential net improvement of
4.5 % in power capture in a study with the NREL 5 MW reference turbine
for the waking wind direction. Finally, a reference turbine which is not
waked in any of the experimental directions is chosen to provide reference
signals. This is turbine R1 in Fig. .
In the initial phase of work, Envision shared with NREL a FAST model of the
turbines FAST is an aero-servo-elastic wind turbine simulation tool
maintained by NREL;. Additionally, Envision provided the
layout of the turbines used in the campaign. Finally, Envision provided
several months of supervisory control and data acquisition (SCADA) data used
to evaluate and tune models in advance of the campaign. As shown in
Fig. , these inputs were then included in the derivation of
the models used in the yaw control strategy, which will be described in the
next section.
ModelingSOWFA
The first model to be used in this study was the SOWFA
model . SOWFA is a wind farm simulation tool,
which models the atmospheric boundary layer using CFD and then models the
turbines using embedded FAST models (in the version of SOWFA used in this
work). The turbines and flow interact through a two-way actuator line
coupling.
SOWFA has been validated against wind farm SCADA data see, for example,. However, validating SOWFA is not the focus
of this campaign. The primary use of SOWFA in this study is generating data
sets of wakes simulated from the Envision turbine, which can be used to tune
the FLORIS model and repeating the procedure that was done for the
NREL 5 MW reference turbine in.
Using the FAST model of the turbine, a suite of simulations is assembled. A
wind condition of 8 m s-1 wind and 6 % turbulence was used for this study. Simulations were then run
with a single turbine operating with various amounts of yaw misalignment, as
well as for two turbine cases, wherein the upstream turbine has various yaw
misalignments and the downstream turbine is placed in various positions
downstream and cross-stream. The trends in power for the single-turbine case
and two-turbine case provide important data with which to tune the
control-oriented FLORIS model, described in the next section.
FLORIS
Using a FAST model of the turbine and the data sets produced by SOWFA, it is
possible to obtain a FLORIS model that can predict the power of turbines in a
farm in steady state, including wake redirection. As discussed in
, the FLORIS model is primarily based on the Jensen
model and the Jiménez model . In particular, FLORIS identifies three
different wake zones with separate wake recovery parameters to capture the
wake characteristics. In addition, FLORIS includes the Jiménez model to
incorporate the wake deflection caused by yaw misalignment. FLORIS contains a
set of parameters to be tuned for a given turbine, including parameters
describing the turbine and wake behavior.
Some parameters of the FLORIS model can be set directly using the FAST model.
Examples of this include the rotor radius and the table of power and thrust
coefficients by wind speed, which can be obtained by running steady wind
simulations in FAST.
The remaining parameters are tuned in this work through an optimization
routine that minimizes the error between the power outputs simulated in SOWFA
and those predicted by FLORIS. As a design rule, the smallest set of
parameters should be adjusted away from their default settings that produce a
reasonable fit. Through iteration, it was determined that adjustments to four
parameters gave a good approximation. The parameters that we focus on are
described below.
The parameter pP relates yaw misalignment to reduction in power by
Lyaw=cos(γ)pP,
where γ is the yaw misalignment of the turbine and Lyaw is
the fraction of power relative to a non-yawed baseline. The higher pP is,
the more quickly a turbine loses power by misalignment and the less likely it
is that wake steering will work because the downstream turbine needs to recover more
power to compensate for power lost by the upstream turbine.
The parameter ke determines the rate at which a wake expands and
recovers to the free-stream velocity. A larger ke value indicates
a faster wake recovery to free stream. Standard values in the literature range
from 0.05 to 0.1. Similarly, kd, based on the Jiménez
model, describes the rate at which a deflected wake reverts to the
free-stream direction. A larger kd parameter indicates that the
wake is less sensitive to yaw misalignment. Standard values in the literature
range from kd=0.1 to kd=0.3. Finally,
WDinit describes an initial wake deflection angle without
steering, which is important for capturing the asymmetry of wake steering.
This asymmetry is likely caused by the combination of the rotation of the
turbine and the shear layer in the atmospheric boundary layer.
The results of the tuning optimization are presented in
Table , which shows both the default values obtained from
tuning to the NREL 5 MW reference turbine as well as the newly obtained
values.
Among these, the lower value of pP is interesting, as 1.43 is below the
value obtained for the NREL 5 MW turbine (1.88) and other experimental
results. For example, the coefficient is fit to wind tunnel tests in
to be 2.0. However, 1.43 is an attractive number
as it implies that wake steering can be performed with less losses incurred
on the upstream yawed turbine.
FLORIS optimal yaw misalignment results. The dashed vertical line
indicates the direction in which the downstream turbine is fully waked by
turbine C1. See Fig. for details on the precise wind
directions. Specifically, the first dashed line (from left to right) refers
to the direction of turbine C1 fully waking downstream turbine D1. The second
dashed line refers to the direction of turbine C1 fully waking downstream
turbine D2. Finally, the third dashed line refers to the direction of turbine
C1 fully waking downstream turbine D3. Wakes from turbines other than C1 are
also evident from the dashed lines. Note that for the power subplots, the
power is given normalized to the power of the turbine in un-yawed and
un-waked conditions
Control design and field test
Given a completed FLORIS model, it is now possible to derive a set of yaw
misalignments for turbine C1 that will optimize power for the pairs of
turbines (D1, D2, and D3 downstream) by wind direction. Wind speed is not
used as an input as inspection indicated minimal sensitivity. It is important
to note that there is not much benefit at very low and very high wind speeds,
suggesting that it is sufficient to enable and disable the controller by wind
speed rather than scheduling.
Some constraints were placed on the optimization. First, for turbine loading
and safety reasons, the maximum yaw misalignment was limited to 25∘.
Second, it was decided for this experiment to limit the controller to
positive yaw misalignment angles (in our nomenclature this is rotating the
turbine counterclockwise from the wind when viewed from above). This is
because this has been demonstrated to be more effective see, for example,where positive yaw misalignments yield higher power increases as
compared to negative yaw misalignments, and including
negative misalignments might raise loads and require a nontrivial transition
from positive to negative yaw misalignments near the wake crossover point,
i.e., when it becomes more beneficial to redirect the wake from the right
side of the downstream turbine to the left side of the downstream turbine.
Using the tuned FLORIS model and these constraints, it is possible to now
derive an optimal table of yaw misalignments for turbine C1. This is shown in
Fig. .
Cosine exponent fit. Panel (a) shows the range of data over
various yaw misalignments and the power ratio. The blue line indicates a
cosine function with the fit exponent value of pP=1.41, and the banded
region indicates the confidence interval as described in the text. Panel (b) indicates the number of points included in each yaw alignment
position.
The optimal yaw misalignment angles for turbine C1 are shown in the upper
left of Fig. . The dashed lines indicate the directions in
which turbine C1 wakes one of the three downstream turbines (see
Fig. ). Near the fully waked directions, the misalignments are
largest and taper down as less deflection is needed to remove the partial
wake overlap situations. The power loss of turbine C1, shown in normalized
power, is indicated in the middle left plot. The overall “plant” gain for
these four turbines is then shown in the lower left. Finally, the right
panels show the normalized power of the three downstream turbines with and
without wake steering. Based on the percent improvement, the sum power is
expected to increase for all three pairings, meaning the gains downstream
exceed the losses upstream. However, this is least so for turbine D3 as at
14.3 D, the baseline wake loss is much less, which is expected.
Using this table of offsets by wind direction, engineers at Envision modified
the yaw controller of turbine C1 to deliver these offsets. Note that this offset
tracking must be done within the limits of yaw control actuation, and for
safety reasons the offset was disabled in sustained winds above
10 m s-1. We note here also that the purpose of the experiment was to
demonstrate the principle of wake steering and not produce a fully optimized
closed-loop control implementation, which is expected to be the subject of
future work. As will be discussed, the present controller offsets correctly
in the average sense, with a wide variation of offsets occurring dynamically.
Following the implementation of the controller, the experimental campaign was
then run in two phases. In the first phase, the wind farm was operated
normally while data relevant for this campaign were collected. This phase
lasted 4 months, from 3 April 2016 to 5 August 2016. The second phase used
the controller designed above and was run an additional 4 months, from
5 August 2016 to 2 December 2016.
Statistical analysis of the SCADA vane data from turbine C1, with
the offset controller off and on. This graph shows the controller of turbine
C1 operating to redirect the wake away from the downstream turbines. The
black and red lines indicate the average yaw offset positions from SCADA data
for the baseline and offset strategy, while the dashed lines are those of the
look-up table (LUT) from FLORIS. Finally, the red and black regions show the
range of points from the 25th through 75th percentiles for the baseline and
offset SCADA data.
It is regrettable that the campaign could not be run longer. Naturally, only
a portion of the data features the wind directions of primary interest.
Additionally, it should be noted that it would be better to alternate the
controller on and off regularly throughout the campaign to better compare
conditions when the controller is on than when it is off. However, Rudong is
a commercial wind site with internal and external restrictions, and it was
necessary to run the test within these constraints. It is important to note
that there are limitations with field testing and data collection. In
particular, the results are impacted because of the relatively short window
of data collection. Further, the sequential testing pattern opens the
possibility of confounding influences such as seasonal variation in
atmospheric conditions. The results of these limitations will be discussed
more in the analysis section.
Results and analysis
In this section, the results of the campaign are presented and analyzed. As a
first step, the behavior of the upstream turbine can be analyzed
independently. Specifically, the pP term, which describes the loss of power
against yaw misalignment, can be derived from the experimental data and
compared with the value computed from SOWFA and used in FLORIS.
The data are collected at a 20 Hz rate from the campaign and are reduced to
1 min averages. Next, we define a power ratio, which is the power of turbine
C1 (the controlled turbine) divided by the power of turbine R1 (the reference
turbine). We then consider this ratio as a function of the yaw misalignment
of turbine C1. This misalignment is measured by the wind vane of turbine C1.
Note that this calculation was repeated, wherein the misalignment is the
difference between the yaw angle of turbine R1 and C1, without substantially
affecting the results. The value of pP is determined as the result of a
minimization problem between the data and Eq. (1). Additionally, to provide a
form of confidence interval, a separate pP is similarly derived for every
day of data, and the 25th and 75th percentile values are used as a range of
confidence. The data, fit, confidence interval, and amount of available data are shown in Fig. .
This figure shows the results fitting power curves for the turbine
C1 to turbine D1 pair. FLORIS-BASE refers to the results from the FLORIS
model with no wake steering, and FLORIS-OPT refers to the results from the
FLORIS model with the optimal yaw set points. SCADA-BASE refers to the field
test where the controller was off, and SCADA-OPT refers to the field test
where the controller was on. The banded region shows the range of fit values from
the 25th to 75th percentile of fitted power curves. In particular,
(a) shows turbine C1 and the resulting power with the controller on
(red) and off (black). Panel (b) shows the downstream turbine, i.e.,
turbine D1, with the offset controller of C1 on (red) and off (black).
Panel (c) shows the overall power between the pair of turbines. There is a
significant power gain when the controller is on versus when it is off.
Finally, (d) indicates the number of days used to compute the
statistics provided in the top three plots.
Using this method, the determined value of pP was 1.41, which is close to
the value derived from SOWFA. This is the first encouraging result, as the
value from SOWFA initially appeared lower than expected. From a wake-steering
perspective, it is useful that the upstream turbine will lose less power when
operating in yawed conditions with a smaller exponent.
Figure shows the performance of the controller in obtaining
the desired offsets. In the figure, for the baseline and optimized control
strategies, the mean offset of turbine C1 is shown, as well as a shaded
region including points (1 min averages) from the 25th through 75th
percentile of observed points. Finally, the initially prescribed pattern of
FLORIS is shown. What can be observed is that implementing the offset
strategy in real conditions where the turbine must track an ever-changing
wind direction within the bounds of the yaw controller results in the sharp
narrow peak of the optimal strategy being smoothed and spread. This implies
that the actual control is somewhere between the baseline and optimum in
terms of performance.
Finally, Figs. , , and
compare the overall performance across the directions for the three
two-turbine pairings. To compute this performance, the following procedure is
used. First, for each wind direction (in 5∘ bins) and for each
turbine a power curve of the form
P=min(aNv3,Prated)
is computed for each day using an error-minimization technique, where a is the fitted
value and N is the nominal value, which includes the air density,
coefficient of power, rotor area, and efficiency losses. v is the wind
speed measured by the reference turbine R1.
The value a represents a scalar gain on the power curve below rated and
also effectively shifts the rated wind speed. N is selected such that a
turbine operating normally and un-waked would on average have an a value of
1.0.
This method, in which the data are used to derive a reduction value to the
power curve, was selected after significant effort to combine the available
data into a complete analysis. Note that alternative methods, such as
directly comparing the power production of the turbines was performed and gives
similar results. The power curves are fit on a per-day basis (and not for
example globally or through random boot-strapping) to help visualize the
day-to-day variation in performance.
The computed value a is the amount by which the power curve is reduced
below nominal as observed for non-waked and non-yawed turbines. From these
values computed for each available day, for each control setting, the 25th,
50th, and 75th percentile values are determined and used as the middle fit
and confidence range. The reason for not computing the power curve globally
was to limit the impact of particular outlier days and give some indication
of the range of results observed. Where the range is small and the number of
days of data collected large, the trends converge. Finally, the power
predictions of FLORIS, for the same conditions, are overlaid on the plots for
comparison. The power values from FLORIS were computed by simulating the same
conditions and then using the same fitting approach to include some of the
effects such as blurring between wind directions in the 5∘ bins.
This figure shows the results of the turbine C1 to turbine D2 pair.
In particular, (a) shows turbine C1 and the resulting power with the
controller on (red) and off (black). Panel (b) shows the downstream
turbine, i.e., turbine D2, with the controller on (red) and off (black).
Panel (c) shows the overall power between the pair of turbines. Finally,
(d) indicates the number of days used to compute the statistics
provided in the top three plots.
This figure shows the results of the turbine C1 to turbine D3 pair.
In particular, (a) shows turbine C1 and the resulting power with the
controller on (red) and off (black). Panel (b) shows the downstream
turbine, i.e., turbine D3, with the controller on (red) and off (black).
Panel (c) shows the overall power between the pair of turbines. There is
no noticeable power gain with the controller on or off. This is likely
because of the increase in spacing between turbines C1 and D3 (14.3 D) in
comparison to the spacing between turbine C1 and D1 (7 D). Finally,
(d) indicates the number of days used to compute the statistics
provided in the top three plots.
For a first analysis, consider the turbine C1–D1 pair in
Fig. . In particular, the results shown in
Fig. show the results from FLORIS with no wake steering
(FLORIS-BASE, shown as the dashed black line with star symbols), FLORIS with optimal yaw set
points (FLORIS-OPT, shown with the dashed red line with star symbols), the field test with no
wake steering (SCADA-BASE, shown with the solid black line with squares), and the field
test with the optimal yaw set points (SCADA-OPT, shown with the solid red
line with squares). For the upstream turbine C1, it can be noted that no trend in power
change can be strongly observed. This is fairly consistent with the limited
power loss of the lower pP exponent and the fact the achieved yaw
misalignments were, on average, not very large. The results for the 7 D
downstream turbine D1 are encouraging. A clear increase in power production
is observed in the wind directions near the primary wake direction of
-20∘ (or 340∘). In the case of the main wake direction, the
power is increased from 0.59 to 0.76 (an increase of 29 %), which is less
than the gain predicted by FLORIS of approximately 40 %. However, it is
evident that the smearing that results from various wind directions, which
was evident in Fig. , is impacting the results as well. The
overall pattern is again of a less dramatic gain; however, it is spread over
a wider area. This result fits with the analysis reported in
, which observed that the Jensen model (and by
extension Jensen-based models like FLORIS) can predict wind farm power
production more accurately if wind direction measurements are assumed to be
uncertain. Incorporating this uncertainty into the FLORIS model should yield
a better fit in that the wake deficits would appear more widely spread, as
they do in the SCADA data. Incorporating this uncertainty into control design
is the subject of ongoing research.
Turning to the power average of the two turbines, the lack of loss upstream
and a good improvement downstream leads to a better than expected return for
the pair. In interpreting Fig. , it is perhaps useful to
focus as much on the shaded regions (which indicate the region containing the
25th to 75th percentile of fitted power curves) as on the solid lines with squares,
which are the 50th percentile. Where the banded regions overlap least is
where we see a persistent change in performance. Large changes in the line,
for example around 5∘ in Figure b are not yet meaningful as the regions are
completely overlapping. That the banded regions of the two-turbine average
contain significant nonoverlapping regions around the wake control direction
is probably the main positive finding of this paper. It suggests that the power
improvement is consistent.
Finally, Fig. d shows the number of
separate days available for computing the data in each bin. It is useful to
note that these do not indicate full days but that some data were collected
on a given day, which was in this direction, and a separate power curve
computed. A value of 10, for example, indicates that 10 days were used for a
given bin for a given controller (sharing no data points), and these values
were used to produce the statistics shown in the plot.
In the wind directions in which wake control is not active (to the left and
right of the plot), there is no persistent trend between the baseline and
optimized control on the downstream turbine. This is also mostly true where
the turbine is waked by the reference turbine R1 at -47∘. This
helps to confirm that the change observed is caused by wake deflection,
rather than a difference in atmosphere between testing periods being the
cause of the underlying changes in wake behavior. This observation could be
made for turbine D2 and 80∘ in Fig. , when it is waked
by turbine O2 (refer to Fig. ), and turbine D3 at 55∘ in
Fig. , when it is waked by reference turbine R1. This last
case is especially compelling, as the spacing is 8.5 D, which is similar to
the space between C1 and D2 and there is a comparable amount of data.
However, unlike D1 and D2 when they are behind the controlled turbine C1, there is no
improvement in power production.
For the pairing of turbine C1 and D2 in Fig. , the spacing
is now 8.5 D and wake steering is expected to become more challenging. Nevertheless,
the combination yields an improvement in the main wake direction
(50∘) for the downstream turbine and the two as a pair. When stepping
away from the main wake direction of 50∘ in either direction,
however, things become ambiguous. To the left, we see the amount of data for
the baseline case grows smaller and the spread in results for the downstream
turbine D2 grows larger (observing the large gray regions). The power is low
despite no wake (although this is a wind direction in which the inflow to
turbine D2 runs in between turbine R1 and turbine C1.) Therefore, it is
probable that a lack of data plays a part, and we would expect little change
here. To the right, the trend in power goes negative for the downstream
turbine and the pair. Nevertheless, it should be noted that the spread of results,
indicated by the bands, is completely overlapping, and so the significance
of this impact cannot be established.
Finally, observing the results of the turbine C1–D3 pair in
Fig. , at a spacing of 14 D, little improvement is expected
and basically none is observed. It is useful to note that turbine D3 has the
most data collected and the results seem best converged. As noted earlier,
when turbine D3 is waked by turbine R1 at 54∘, it is the deeper wake,
having a spacing of 8.5 D, and no noticeable change in wake loss occurs (the power of
the downstream turbine is the same in the baseline and optimized cases); this
points to wake steering being the primary cause of change in power for the
earlier cases at 7 and 8.5 D between the baseline and optimized cases.
Also, it is interesting to consider what is happening west of 80∘
for the downstream turbine D3. FLORIS predicts a return to full power,
followed by a dip around 95∘ when turbine O2 is upstream. However,
what is actually observed is a reduction in power basically across the whole
range. Considering Fig. , this is a range of wind direction
without an obvious single-turbine wake but with four turbines still
upstream. Notice that unlike an explanation of shallower but more spread loss
from wind direction uncertainty, this deficit is deeper than what is
predicted by FLORIS. This deficit speaks to an unmodeled deep array effect
that may prove important to include when FLORIS is used to model and design
controllers for multi-turbine arrays.
Conclusions and future work
This study provided several encouraging, albeit qualified results. The main
result was that for the directions and spacings (7 D and 8.5 D) expected to
produce an improvement in power for the pair of turbines, such an improvement
is observed. The most easily interpreted results come from the closer 7 D spacing,
whereas the 8.5 D spacing has some changes that are partly caused by limited
data availability. Another good result was the observed agreement between the
lower than expected power loss with yaw function predicted by SOWFA and the
loss derived experimentally. This result is positive because it provides
another form of validation for SOWFA (data sets of utility turbines operating
misaligned are not commonly available for testing), and the low power loss
value makes wake steering in general more successful.
In this paper, it was discussed that the primary limitations were constraints
placed on the amount of data that could be collected at a commercial wind
farm and constraints placed on the wind turbine yaw controller's ability to
control. In the case of the pairing with turbine D3, the control and optimal
cases are very close in midpoint and range, suggesting that 30–40 separate
days of testing is a good target per controller. Had it been possible to
toggle between control set points, it is very likely that this number could
be reduced although toggling creates some issues of transition. On the
controller side, it would appear that the implemented controller was
sufficient to secure power gains; however, it could be that a more advanced
controller could achieve more.
Also an important concern is the impact this control will have on turbine
loads, both of upstream and downstream turbines. This question was outside
the scope of the present campaign, and the turbine load data were not
available for this joint study. However, there are ongoing research campaigns
which seek to understand and quantify the impacts to loading. Already
published are studies which consider the impact on loads from operating in
yaw using aero-servo-elastic wind turbine models and
computational fluid dynamics . Additionally, there is a
currently ongoing field test in which a utility-scale wind turbine which is
highly instrumented with load sensors is operated intentionally in yaw to
assess impacts . Finally, in order to systematically
understand the impact on loads of turbines downstream from wake steering, new
engineering tools are being developed, which include mid-fidelity models of
wakes, such that loads can be simulated, but which are computationally inexpensive
enough for load suites to be computed .
More generally, the design of closed-loop control systems for wake steering
remains an open topic of research. The present method can be regarded as open
loop because the actual wake to be controlled is never observed or estimated
by the controller. Research that uses lidar to track and control wakes
or uses estimation techniques
may very well improve upon these first results. This is also the subject of
ongoing multiyear research projects in the United States and
Europe .
No data sets were used in this article.
The authors declare that they have no conflict of
interest.
Acknowledgements
The authors would like to acknowledge and thank Matthew Churchfield for his
support in the use of SOWFA and the development of the simulations.
Additionally, the authors thank Pieter Gebraad for his work on this effort
during his tenure at NREL.
The US Government retains, and the publisher, by accepting the article for
publication, acknowledges that the US Government retains, a nonexclusive,
paid-up, irrevocable, worldwide license to publish or reproduce the published
form of this work or to allow others to do so, for US Government purposes.
Edited by: S. Aubrun Reviewed
by: two anonymous referees
ReferencesA2e: Atmosphere To Electrons,
http://energy.gov/eere/wind/atmosphere-electrons, last access: 15 December 2016.
Campagnolo, F. P., Vlaho Bottasso, C. L., and Croce, A.: Wind tunnel testing of wake
control strategies, in: American Control Conference (ACC),
Churchfield, M., Lee, S., Moriarty, P., Martinez, L., Leonardi, S.,
Vijayakumar, G., and Brasseur, J.: A large-eddy simulation of wind-plant
aerodynamics, in: 50th AIAA Aerospace Sciences Meeting Including the New
Horizons Forum and Aerospace Exposition, Nashville, TN, Nashville, Tennessee,
USA, 2012. American Automatic Control Council (AACC), 513–518, 2016.Churchfield, M. and Lee, S.: NWTC Information Portal (SOWFA),
https://nwtc.nrel.gov/SOWFA (last access: 1 February 2017), 2014.CL-Windcon: CL-Windcon project kicked-off in Pampolna,
http://www.ifb.uni-stuttgart.de/news/CL-Windcon-Kickoff, last access:
15 December 2016.
Doekemeijer, B. M.: Enhanced Kalman filtering for a 2D CFD Navier-Stokes
wind
farm model, Ph.D. thesis, Delft University of Technology, 2016.
Fleming, P. A., Gebraad, P. M. O., Lee, S., van Wingerden, J. W., Johnson,
K., Churchfield, M., Michalakes, J., Spalart, P., and Moriarty, P.:
Simulation
comparison of wake mitigation control strategies for a two-turbine case, Wind
Energy, 18, 2135–2143, 2014a.Fleming, P. A., Gebraad, P. M., Lee, S., van Wingerden, J.-W., Johnson, K.,
Churchfield, M., Michalakes, J., Spalart, P., and Moriarty, P.: Evaluating
techniques for redirecting turbine wakes using {SOWFA}, Renew. Energ.,
70, 211–218, 10.1016/j.renene.2014.02.015, 2014b.
Fleming, P. A., Ning, A., Gebraad, P. M. O., and Dykes, K.: Wind plant system
engineering through optimization of layout and yaw control, Wind Energy,
19, 329–344,
2015.Fleming, P., Churchfield, M., Scholbrock, A., Clifton, A., Schreck, S.,
Johnson, K., Wright, A., Gebraad, P., Annoni, J., Naughton, B., Berg, J., Herges, T., White, J., Mikkelsen, T., Sjoholm, M., and Angelou, N.:
Detailed field test of yaw-based wake steering, J. Phys.
Conf. Ser., 753, 052003, 10.1088/1742-6596/753/5/052003, 2016.
Fleming, P., Annoni, J., Scholbrock, A., Quon, E., Dana, S., Schreck, S.,
Raach, S., Haizmann, F., and Schlipf, D.: Full-Scale Field Test of Wake
Steering (Under Review), in: Wake Conference, Visby, Sweden, 2017.
Gaumond, M., Réthoré, P.-E., Ott, S., Peña, A., Bechmann, A., and
Hansen, K. S.: Evaluation of the wind direction uncertainty and its impact on
wake modeling at the Horns Rev offshore wind farm, Wind Energy, 17, 1169–1178, 2013.
Gebraad, P., Teeuwisse, F., van Wingerden, J., Fleming, P., Ruben, S.,
Marden,
J., and Pao, L.: Wind plant power optimization through yaw control using a
parametric model for wake effects – a CFD simulation study, Wind
Energy, 19, 95–114, 2014.Gebraad, P., Thomas, J. J., Ning, A., Fleming, P., and Dykes, K.:
Maximization
of the annual energy production of wind power plants by optimization of
layout and yaw-based wake control, Wind Energy, 20, 97–107, 10.1002/we.1993,
2016.
Jensen, N. O.: A note on wind generator interaction, Tech. Rep. Risø-M-2411,
Risø National Laboratory, 1984.
Jiménez, Á., Crespo, A., and Migoya, E.: Application of a LES
technique to characterize the wake deflection of a wind turbine in yaw, Wind
Energy, 13, 559–572, 2010.
Jonkman, J. M. and Buhl Jr., M. L.: FAST Manual User's Guide, NREL report
No. NREL/EL-500-38230, 2005.
Jonkman, J. M., Annoni, J., Hayman, G., Jonkman, B., and Purkayastha, A.:
Development of FAST, Farm: A New Multi-Physics Engineering Tool for Wind-Farm
Design and Analysis, in: 35th Wind Energy Symposium, 0454, 2017.
Medici, D.: Experimental studies of wind turbine wakes: power optimisation
and
meandering, Ph.D. thesis, KTH, Mechanics, qC 20101018, 2005.
Raach, S., Schlipf, D., Borisade, F., and Cheng, P. W.: Wake redirecting
using
feedback control to improve the power output of wind farms, in: American
Control Conference (ACC), American Automatic Control
Council (AACC), 1387–1392, 2016.
Schottler, J., Hölling, A., Peinke, J., and Hölling, M.: Wind tunnel
tests on controllable model wind turbines in yaw, in: 34th Wind Energy
Symposium, 1523, 2016.
Schulz, C., Letzgus, P., Lutz, T., and Krämer, E.: CFD study on the
impact of yawed inflow on loads, power and near wake of a generic wind
turbine, Wind Energy, 20, 253–268, 2016.Trujillo, J. J., Seifert, J. K., Würth, I., Schlipf, D., and Kühn,
M.: Full-field assessment of wind turbine near-wake deviation in relation to
yaw misalignment, Wind Energ. Sci., 1, 41–53, 10.5194/wes-1-41-2016,
2016.Vollmer, L., Steinfeld, G., Heinemann, D., and Kühn, M.: Estimating the
wake deflection downstream of a wind turbine in different atmospheric
stabilities: an LES study, Wind Energ. Sci., 1, 129–141,
10.5194/wes-1-129-2016, 2016.
Wagenaar, J. W., Machielse, L., and Schepers, J.: Controlling wind in ECN's
scaled wind farm, in: EWEA Annual Meeting, Copenhagen, Denmark, 2012.
Zalkind, D. S. and Pao, L. Y.: The Fatigue Loading Effects of Yaw Control
for
Wind Plants, in: American Control Conference, Boston, MA, 2016.